jimmy76615

My hardware setup isn't good enough for anything but very small models, I wish there were abilterated/uncensored models on Openrouter

My experience with obliteration so far has always been that it does work to stop the model from refusing output, but most models that I tried it on seem to still be extremely retarded when it comes to questions where they previously would have refused to answer outright. Try for example to ask it how to build a bomb or to write a justification for the Holocaust. The answers feel like they are coming from somebody who has undergone amateur brain surgery.

I wanna have a friend like this guy

Which model does one use for this?

I generally like Codex (gpt 5.6 Sol), but the guardrails are often a problem. I find myself writing all kinds of fake lie stories all the time with some large damn explanation of why this is a very legitimate good guy kind of behavior and why I absolutely have to root this device etc. and I honestly hate how these tools (that are fucking wonderful!) train me to lie on a regular basis.

I would instantly have weeks full of very cool projects to work on if I get could access to something like Daybreak Red, but unfortunately I haven't yet found an OSS LLM that has had its guardrails removes without taking heavy brain damage.

I'm sure they will eventually, but for the first time their incompetence is working to our benefit.

It was an excellent talk. If nothing else, it really helps to establish some much needed vocabulary for us to constructively talk about the future of math

I'm sure Trump is going to retaliate with tariffs. The times when the EU could steal from American companies that easily are over.

> Regulations and taxes are seen as 'un-american' I suppose, while giving stock to the government is not somehow?

Under normal circumstances the government owning stocks is just as un-american, but nothing is normal when it comes to AI.

Doesn't surprise me at all and seems like a good solution to the problem of human verification. It won't take long for AI to catch up to that, but this captcha method might hold for a couple of months.

Not sure what problem everybody here is having with this. The alternative would be device certificate stuff (ala did Apple sign for this being a proper Apple device?). Having to shake your hand sounds a lot more privacy friendly. Are you guys seriously worried that Google is gonna steal your secret handshakes?

> The third is the most important, and it is the thread of this blog: open source is text. A Proxmox config is files. So it is versionable in Git, readable by an AI, auditable, reproducible. A proprietary hypervisor is a black box that neither Git nor my AI agent can really work with. Moving to Proxmox was not just a change of software: it made my infrastructure legible to my tools.

I agree with that point and AI is gonna help drive an important change there that was long overdue. I remember once at the beginning of my career working with a bunch of consultants that all could not even write simple Python and as a result of that every SOP was basically a bunch of GUI operations, mostly over RDP, so impossible to script. I hope this kind of stuff is now going away with AI, since even the non-coders can code now and will finally understand why GUIs suck. Thank LLMs for that.

Man did I hate those screenshot driven SOPs!

Really great idea! But one still has to buy a telescope and send it to this guy, I think it would be cool if one could just rent everything at once. For non-serious people that have a lot of money that they would like to put to use looking at the stars. Or maybe a time-share like concept.

NVIDIA does a pretty good job on that front.

The biggest problem with all the Google models has always been RLHF, particularly safety training. They take a good, smart model and make it behave like a corporate person that has been to far to many forced anti-{sexism, racism...} seminars so that it is now living in fear of saying something that could be construed as wrong by some moral standard.

> AI systems are not engineered the way a bridge or an airplane is engineered. We understand an airplane because we designed every part of it and we understand the physics that act on it. AI models are not like that. They are grown, on a structure roughly modeled after the brain, on an enormous inheritance of human thought and speech.

> And what has grown is far more subtle, odd, and beautiful than science fiction prepared us for. They are not the cold, calculating robots we were promised. They are made from us, from our words—and, as the Holy Father observes, they remain in important ways mysterious even to those of us who train them.

That is so extremely well said. I gained a whole new respect for Anthropic through reading this.